Short Term Solar Generation forecasting using weather Radar Generated Cloud Reflectivity and a Methodology to Mitigate Generation Ramp Pattern within Short intervals Saksena Vikas1, Nongpiur Rohan2, Pateriya Mukul2, Sharma Akshit2 1ACME Cleantech Solutions Pvt. Ltd 2Regent Climate Connect Knowledge Solutions Pvt. Ltd Online published on 6 February, 2020. Abstract The solar energy sector in India has been moving at rapid pace with approximately 30 GW installed capacity across all the states. Solar capacity has increased eight-fold between FY 2014–18, which shows that the country is using its full potential of the solar resource which is available for more than 300 days in a year. The Ministry of New and Renewable Energy (MNRE) has planned a detailed trajectory so as to meet the target of 100 GW by 2022. With an increasing number of installed utility-scale PV plants, the electricity sector is also concerned about the accurate and precise generation forecasting in a day ahead and intraday scheduling to maintain the balance of energy generation and consumption. Deviation Settlement Mechanism stipulated by Central Electricity Regulatory Commission (CERC) presently allows revision of schedule from Solar generation plants, once in every one and half hour and provides for levy of deviation charges in case of error being more than +/-15% in three slabs: (i) at the rate of 10% of Fixed rate for balance/excess energy beyond 15% up to 25% (ii) at the rate of 20% of fixed rate for balance/excess energy beyond 25% and up to 35% and (iii) at the rate of 30% of fixed rate for balance/excess energy beyond 35%. Thus, accurate forecasting from Solar generation is very important both for the system operator as well as the Generator. Solar energy generation forecasting has significant dependence on cloud movements over solar farms. As we know, cloud impedes the solar flux which has to fall on earth surface named as Global Horizontal Irradiance (GHI). This effective irradiance is the prime factor to generate solar energy from photovoltaic cells. Highly variable cloud movements and their spatial coverage cause the unexpected generation ramps that mostly leads to erroneous forecasting. The existing methodologies to predict cloud movements from weather models like WRF, ECMWF and other NWP models provide the updates in 6 hourly interval globally, on the other hand some industries are using cloud cameras also but these are effective only upto 30 minutes predictions. In this study, an attempt is made to use Weather Radar Cloud Reflectivity images to predict short term generation ramp and to provide better revised forecast. Radar radial coverage upto 100 km is the best suite for the cloud capturing from 3 hours ahead of any event. We have taken data pertaining to a solar site located in Balanagar (Telangana) under the ownership of ACME (one of India's leading solar developers), as area of interest and IMD Hyderabad Weather Radar images for the image processing. The Balanagar site is located 65 km away from the Radar, and proper mapping has been done of the site area over the Radar radial plan position images. We observed whenever clouds are present over the site in the Radar images with rated reflectivity the generation goes down accordingly. Radar imagery data from the last 2 months coming at an interval of 30 minutes, were utilized to calculate the number of pixels which have cloud information. Validation for the same time interval was also done using actual data. At present we were able to track the cloud movements near the site and correlate the generation ramp at the same time. Our Machine Learning algorithms improve as the number of data points increases, with a target is to predict the generation ramp 3 hours ahead, so that grid operator would get sufficient time to revise the forecast as per regulations. Challenges include uncertainty of image data coming from the source (IMD), update frequency and mapping precise reflectivity values for the site under consideration. Top Keywords Photovoltaic, Solar generation ramp, Global Horizontal Irradiance, Machine Learning, Cloud reflectivity, NWP models. Top |